Comprehensive analysis of leak impacts on liquid-gas multiphase flow using statistical, wavelet transform, and machine learning approaches

Abstract

Detecting small, subtle, and closely spaced leaks is considerably more challenging than identifying larger leaks, particularly under multiphase flow conditions. The inability of current models to consistently detect small leaks or distinguish between multiple leaks and a single leak highlights the need for enhanced detection techniques. Although pressure responses over time for single and multiple leaks are highly similar, additional analyses such as frequency analysis, wavelet analysis, and artificial intelligence can distinguish between these scenarios. In this study, experimental tests were performed on a horizontal flow loop system with a diameter of 50.8 mm equipped with three controlled artificial leaks in the middle section of the pipeline. Statistical, Wavelet Transform (WT), and Machine Learning (ML) approaches were applied to the recorded time-series signals (dynamic pressure) for various operating conditions of liquid and gas superficial velocities. Our findings demonstrate that these additional analyses can effectively distinguish between single-leak, multiple-leak, and no-leak scenarios. Additionally, the impact of leaks on the flow regime map in a pipeline was discussed. The revealed results could offer novel perspectives regarding process safety and risk engineering including the impact of leaks on multiphase flow systems and their identification.